From a0c81f8549c9a8c5c6c2b8c4e8bded2adeefe3b0 Mon Sep 17 00:00:00 2001
From: vishnuprakaz <nvishnuprakash02@gmail.com>
Date: Thu, 30 Jul 2026 15:31:09 +0530
Subject: [PATCH] Parquet: Guard ordered predicates on variant/nested types in
 metrics row group filter

---
 .../data/TestMetricsRowGroupFilter.java       | 56 +++++++++++++++++++
 .../parquet/ParquetMetricsRowGroupFilter.java | 28 ++++++++++
 2 files changed, 84 insertions(+)

diff --git a/data/src/test/java/org/apache/iceberg/data/TestMetricsRowGroupFilter.java b/data/src/test/java/org/apache/iceberg/data/TestMetricsRowGroupFilter.java
index edb01d7e77ef..a8e19f32492e 100644
--- a/data/src/test/java/org/apache/iceberg/data/TestMetricsRowGroupFilter.java
+++ b/data/src/test/java/org/apache/iceberg/data/TestMetricsRowGroupFilter.java
@@ -1259,6 +1259,62 @@ public void testVariantFieldIn() throws IOException {
         .isTrue();
   }
 
+  @TestTemplate
+  public void testVariantFieldLt() throws IOException {
+    assumeThat(format).isEqualTo(FileFormat.PARQUET);
+
+    VariantMetadata md = Variants.metadata("k");
+    Variant v0 = createVariantWithKey(md, "v0");
+    List<GenericRecord> records = createVariantRecords(v0);
+
+    boolean shouldRead = shouldReadVariant(lessThan("variant_field", v0), records);
+    assertThat(shouldRead)
+        .as("Should read: variant lt filters must be evaluated post scan")
+        .isTrue();
+  }
+
+  @TestTemplate
+  public void testVariantFieldLtEq() throws IOException {
+    assumeThat(format).isEqualTo(FileFormat.PARQUET);
+
+    VariantMetadata md = Variants.metadata("k");
+    Variant v0 = createVariantWithKey(md, "v0");
+    List<GenericRecord> records = createVariantRecords(v0);
+
+    boolean shouldRead = shouldReadVariant(lessThanOrEqual("variant_field", v0), records);
+    assertThat(shouldRead)
+        .as("Should read: variant ltEq filters must be evaluated post scan")
+        .isTrue();
+  }
+
+  @TestTemplate
+  public void testVariantFieldGt() throws IOException {
+    assumeThat(format).isEqualTo(FileFormat.PARQUET);
+
+    VariantMetadata md = Variants.metadata("k");
+    Variant v0 = createVariantWithKey(md, "v0");
+    List<GenericRecord> records = createVariantRecords(v0);
+
+    boolean shouldRead = shouldReadVariant(greaterThan("variant_field", v0), records);
+    assertThat(shouldRead)
+        .as("Should read: variant gt filters must be evaluated post scan")
+        .isTrue();
+  }
+
+  @TestTemplate
+  public void testVariantFieldGtEq() throws IOException {
+    assumeThat(format).isEqualTo(FileFormat.PARQUET);
+
+    VariantMetadata md = Variants.metadata("k");
+    Variant v0 = createVariantWithKey(md, "v0");
+    List<GenericRecord> records = createVariantRecords(v0);
+
+    boolean shouldRead = shouldReadVariant(greaterThanOrEqual("variant_field", v0), records);
+    assertThat(shouldRead)
+        .as("Should read: variant gtEq filters must be evaluated post scan")
+        .isTrue();
+  }
+
   @TestTemplate
   public void testUUID() {
     assumeThat(format).as("Only valid for Parquet").isEqualTo(FileFormat.PARQUET);
diff --git a/parquet/src/main/java/org/apache/iceberg/parquet/ParquetMetricsRowGroupFilter.java b/parquet/src/main/java/org/apache/iceberg/parquet/ParquetMetricsRowGroupFilter.java
index 0cf4aa2a3633..42b06f7c1378 100644
--- a/parquet/src/main/java/org/apache/iceberg/parquet/ParquetMetricsRowGroupFilter.java
+++ b/parquet/src/main/java/org/apache/iceberg/parquet/ParquetMetricsRowGroupFilter.java
@@ -237,6 +237,13 @@ public <T> Boolean notNaN(BoundReference<T> ref) {
     public <T> Boolean lt(BoundReference<T> ref, Literal<T> lit) {
       int id = ref.fieldId();
 
+      // Leave all nested column type and variant type filters to be
+      // evaluated post scan.
+      Type type = schema.findType(id);
+      if (type instanceof Type.NestedType || type.isVariantType()) {
+        return ROWS_MIGHT_MATCH;
+      }
+
       Long valueCount = valueCounts.get(id);
       if (valueCount == null) {
         // the column is not present and is all nulls
@@ -267,6 +274,13 @@ public <T> Boolean lt(BoundReference<T> ref, Literal<T> lit) {
     public <T> Boolean ltEq(BoundReference<T> ref, Literal<T> lit) {
       int id = ref.fieldId();
 
+      // Leave all nested column type and variant type filters to be
+      // evaluated post scan.
+      Type type = schema.findType(id);
+      if (type instanceof Type.NestedType || type.isVariantType()) {
+        return ROWS_MIGHT_MATCH;
+      }
+
       Long valueCount = valueCounts.get(id);
       if (valueCount == null) {
         // the column is not present and is all nulls
@@ -297,6 +311,13 @@ public <T> Boolean ltEq(BoundReference<T> ref, Literal<T> lit) {
     public <T> Boolean gt(BoundReference<T> ref, Literal<T> lit) {
       int id = ref.fieldId();
 
+      // Leave all nested column type and variant type filters to be
+      // evaluated post scan.
+      Type type = schema.findType(id);
+      if (type instanceof Type.NestedType || type.isVariantType()) {
+        return ROWS_MIGHT_MATCH;
+      }
+
       Long valueCount = valueCounts.get(id);
       if (valueCount == null) {
         // the column is not present and is all nulls
@@ -327,6 +348,13 @@ public <T> Boolean gt(BoundReference<T> ref, Literal<T> lit) {
     public <T> Boolean gtEq(BoundReference<T> ref, Literal<T> lit) {
       int id = ref.fieldId();
 
+      // Leave all nested column type and variant type filters to be
+      // evaluated post scan.
+      Type type = schema.findType(id);
+      if (type instanceof Type.NestedType || type.isVariantType()) {
+        return ROWS_MIGHT_MATCH;
+      }
+
       Long valueCount = valueCounts.get(id);
       if (valueCount == null) {
         // the column is not present and is all nulls